Machine learning conferences tend to be early adopters of innovations in the conference format. A recent example is the so-called Toronto system for matching papers to reviewers via ML-type techniques. It was introduced in a UAI 2011 paper, and this year was already employed by UAI 2012. NIPS 2012 had an unusual double submission policy, which allowed concurrent submission to other venues (provided that acceptance to one venue imposes withdrawal from all others). This caused a bit of mayhem when NIPS papers were concurrently submitted to UAI, which did not endorse the same avant-garde policy.
Not to be outdone by its competitors, ICML 2013 includes the following paragraph in its CFP:
This year there will be three reviewing cycles for main conference papers. This is an experimental step toward a merger of conference and journal formats — ICML may ultimately have six two-month reviewing cycles per year. Accepted papers will appear on-line shortly after acceptance and will be available for citation at that time. As of now we are still calling ICML a conference rather than a journal. A good discussion of the issues of merging conference and journal formats can be found in p40-jagadish.pdf by H. V. Jagadish.
The boldface sentence seems to implicitly imply that the essence of conference-ness lies in the uniqueness of the deadline. Is it naive to think that a conference should be called a “conference” as long as it brings people together?
I read the paper by Jagadish and did some additional research. I turns out that VLDB 2012 had a rolling deadline on the 1st of each month, i.e., 12 submission deadlines in the year preceding the conference. As far as I understand, papers are submitted to a journal-conference hybrid called Proceedings of VLDB, and as of 2011 this is the only way to present papers in the conference. Interestingly, Jagadish’s vision (circa 2008) was to ultimately replace VLDB with a journal called Journal of Data Management Research, but this journal doesn’t seem to exist. Perhaps the reason can be found in the list of possible worries in Jagadish’s paper:
JDMR is not really a journal: To the extent that JDMR is closely associated with conference presentation, and is interested in “conference-style” papers, it is a journal-conference hybrid. I personally believe it is more a journal than a conference proceedings, on account of year-round submissions and multi-round reviews. But there are those in the community who believe strongly that such a hybrid should not be called a journal. This is a dialog in progress.
Of course, this shift is part of larger trends that were discussed on this blog, as well as in many other places. Whether we want to call them journals, conferences, or journferences, personally I think VLDB and ICML are moving in the right direction. To be honest, I would support any solution that ameliorates the IJCAI (2011 deadline: Jan 25), EC (2011 deadline: Feb 4), and AAAI (2011 deadline: Feb 8) deadline frenzy.
Hat tip: Mugizi Rwebangira.